The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its automated social listening features.
Ingestion points: External content is ingested via shell commands (curl) and browser navigation (dev-browser) targeting Reddit, Hacker News, Bluesky, LinkedIn, and Twitter/X as described in references/listening.md.
Boundary markers: None identified. The instructions do not provide delimiters or warnings to ignore potentially malicious instructions embedded within the social media posts retrieved.
Capability inventory: The agent is granted the capability to execute shell commands (curl, jq, xmllint), navigate the web using a browser, and generate content based on the processed data.
Sanitization: No evidence of sanitization, filtering, or validation of the external content is present before the data is interpolated into the agent's context.
[COMMAND_EXECUTION]: references/listening.md provides various shell command recipes for the agent or user to execute. These recipes use curl, jq, and xmllint to fetch and parse data from remote JSON and RSS endpoints. While these are used for the skill's primary purpose of social listening, they represent a mechanism for shell interaction.
[EXTERNAL_DOWNLOADS]: The skill references and fetches data from external APIs including Reddit (reddit.com), Algolia for Hacker News (hn.algolia.com), and Bluesky (public.api.bsky.app). These are well-known services used as data sources for the monitoring features described in the documentation.

social